This document contains all the information required to complete the virtual field course. In this course we will carry out a series of virtual point counts using real sound recordings collected from point counts completed high in the Colombian Andes. At each point we will record the bird species that we think are present, and then, once we have completed all the point counts, we will compare how communities vary between pasture and forest points. Because you are unlikely to be able to identify Colombian birds from their song, rather than recording real species, we will instead record “morphospecies”, which we can think of as distinct vocalisations that we think are likely to be made by different species. If we were doing this by eye, rather than by sound, it might be something like “I saw a medium-sized red bird with a heavy beak, a blue and black bird with some yellow on the wing, and a large bright green bird with a big bill”. Without knowing the true species these represent, we can still use these descriptions to record them each time they crop up. Here we will do the same, but with the vocalisations (this will start to make more sense as we work through the recordings).

1 Getting set up

1.1 Download the zip file

This zip file contains all the code and files that you need for this practical, so go ahead and download it now. The practical.zip file you have just downloaded is a .zip file, which is just a way of compressing all the data to create a file with a smaller filesize. Before doing anything else, we need to first unzip it to retain the original files. If you right click on the practical.zip that you have downloaded, you will have a list of options, and you need to select ‘Extract all’. By default it will extract all the files to the directory that the zip file is in: this is fine, but if you would like to save the files somewhere in particular, you can also specify a different directory at this stage. When it has done this you should be able to see a practical folder that, unlike the zipped version, no longer has a zip across the thumbnail.

Important: Make sure you do the unzipping above: you will have problems down the line otherwise. If you aren’t sure, ask!

1.2 Open the template datatable

In the files directory you will fine a template_datatable file, both as a .csv and a .xlsx. Open up whichever one you feel like (they are the same, but the csv version has obviously dropped the excel-specific formatting).

Important: You need to make sure that you don’t alter the formatting of this datafile. There are 11 columns for you to enter information into, and 45 rows (not including the header). There are more rows than there are species in the recordings, so you don’t need to add any rows. You also don’t need to edit the morphospecies column, nor should you remove any rows. It is important that you only put numbers in the point columns. If there isn’t a particular species on a particular point, just leave the cell empty. Try to avoid accidentally putting spaces (" “) in cells you aren’t using, and don’t add any additional columns. You can put whatever you want in the notes column (it might help, for example, to record some notes about the sound of a particular morphospecies, e.g. ”high te-te-te-te“, or”descending series of notes"). If you do accidentally do any of these things, you might struggle to read the data into R.

1.3 Saving your completed datatable

When we have come to the end of the listening section of this practical, you need to save your completed datatable as completed_datatable.csv, in the files directory (the same place you read the template version in from). Note that you have to save it as a .csv file, with the correct name and in the correct location: you won’t be able to read it in if you don’t do this. To save as a .csv, click ‘Save as’, navigate to the files directory, and in the ’Save as type` option, scroll down to ’CSV (Comma delimited) (*.csv)’.

Important: If you aren’t sure how to save the file correctly, just ask a demonstrator. It can be quite tricky to debug your code if you go off-piste at this stage!

2 Sound files

You can listen to the soundfiles here directly, but if you are struggling to play them for some reason, they are also provided in the zip file, in the sound_files directory. We are now going to travel around all of our point counts, listening to each of the recordings and noting the species that we think are present on each point, and how many of them we think there are.

2.1 Practice

2.2 Forest

Point 1

Point 2

Point 3

Point 4

Point 5

2.3 Pasture

Point 6

Point 7

Point 8

Point 9

Point 10

3 Analysing and plotting the data

We are now going to analyse the data: if you require any assistance as we work through this, just ask a TA for help (don’t suffer in silence!). We are going to rely on two packages that you should have already used quite a bit: dplyr and ggplot2. If you haven’t already got these, you will need to install them using install.packages("dplyr") or install.packages("ggplot2").

You need to open the project APS278_practical.Rproj (in the zip file you have already downloaded). You can either navigate to the directory and double click on the project there, or you can open Rstudio and then go File>Open Project and then find the project directory and select the project. Having done that the plotting_script.R file should open automatically, but if it doesn’t just go ahead and open it yourself).

It should hopefully be pretty self-explanatory. The main problem will be if you have not saved your completed data table as completed_datatable.csv, and/or you have not saved it into the files directory (where you read template_datatable from). See script for further debugging tips if you are struggling to read your file in.

Run the code that is already entered and this run some code to format your data, and you can check that your data were read in properly, and inspect the objects that were created for you.

3.1 Your task

  1. calculate the average point-level species richness: on average are there more species in forest or pasture?
    • you already have the summary in data_summary: you want to group_by() and summarise to get an average for each habitat type
  2. plot the point-level species richness. You only have 5 points in each habitat, so a dot plot or jittered points might be the best way to do this
    • geom_jitter() or geom_dotplot() will be useful.
  3. calculate the total species richness in each habitat type
    • data_longformat contains all the information you need for this.
    • You need the number of unique species per habitat. unique(x) will give you a vector of the unique values in vector x, and length(unique(x)) will give you the number of unique values in x.
  4. plot the rarefaction curve for species richness
    • See below for what exactly a rarefaction curve is trying to do. You just need to plot the outputs given to you in the pred_both tibble.
    • geom_ribbon() takes three main aes() arguments, ymin, ymax, and x. You will probably also want to provide it a fill argument so that it plots different ribbons for each habitat type!

Note: It might take you a bit of playing around to figure out how to produce some of these plots/summary statistics. If you are really stuck with something, ask a TA for help. If you aren’t sure about what arguments you need to provide to a plotting function, you can also check the documentation by using ?geom_jitter. The Aesthetics section tells you what arguments you need to provide to it.

3.2 Example figures

These are the plots that I have drawn from my listening to these points. Note that yours will probably look different, depending on how well you have been able to identify species on the recordings. Try to recreate them as best as you can (you can edit the format as you like, pick a colour scheme, etc.)

Note that I have removed the labels on the y-axis as it gives the game away!

3.2.1 Point-level species richness:

3.2.2 Rarefaction curves for each habitat:

What is a rarefaction curve? the rarefaction curve is estimating the species richness you would expect to see in a random sample of \(m\) individuals in a particular habitat. In habitats where there are lots of individuals, but very few species, we expect to see a curve that is very flat (as each additional individual is likely to be from a species we have already seen). At the other extreme, if every new individual is a new species, we expect to see a 1:1 line as the number of species increases in lockstep with the number of individuals. At some point we expect all curves to plateau, as there has to be a finite number of species in any given community. The closer we are to a plateau, the closer we are to having exhaustively sampled the community. Or that’s the idea, at least.

Why do we use them? We will rarely have observed exactly the same number of individuals in different habitat types, and we need a way of making a fair comparison when you have more individuals in one sample than another. The more individuals you have observed, the more chance you have had to spot a new species, so by chance alone you expect to see more species with increasing \(m\). If you want to make a species-richness comparison between two habitats while standardising for the number of individuals, you can use a rarefaction curve. Note that here we have number of individuals on the x-axis, but in other contexts we might consider adjusting for varying surveying effort in other ways, for example by number of points visited, or length of time surveyed, etc.

What is the shaded area? And why the difference between these two figures? The shaded area indicates the variation in average species richnesses that we would expect to see by chance. This is also called the sampling variation in the expected species richness. If we can calculate this for each habitat type (left-hand figure), why have I also produced a plot where this range is only presented for the forest?

If our question is “is the species richness lower/higher in forest than pasture”, we want to know whether we can statistically differentiate between pasture and forest communities. To put this another way: could the observed richness in pasture plausibly have been generated through random sampling of the forest community?

In forest we observed a lot more species (or at least I did!), so the question is: if we had observed the same number of individuals in forest as in pasture, do we observe more/fewer species in pasture than we would expect by chance? To do this, we can treat our species richness in pasture as known- it’s simply what we observed. We can then compare this value to the range of values we would expect to observe through simple sampling variation in forest. If the actual value that we observed lies a very long way away from the range of values that ought to arise through sampling variation then we can conclude that it is very unlikely that our two communities are the same.

Important caveats:

Note that as with any statistical model, the predictions that we make are conditional on the assumptions of the statistical model that we use. So while here we would conclude that the two communities are different (and that there are fewer species in pasture than in forest), this conclusion is conditional on the specific details of this particular model.

We have ignored, for example, that species detection might vary between habitat types, and between points. We have also ignored that each observation is not necessarily independent of another. Imagine for example that a particular species travels around in large flocks: if we observe one individual we are also likely to observe many others! The consequence is that our estimated sampling variation substantially underestimates the true variation that would arise from this overdispersion in abundance. Some of these assumptions don’t seem very reasonable, so before we get carried away and start writing our Nature paper, it’s important that we acknowledge that this is a substantially oversimplified model of reality. Even if we feel that these assumptions are likely reasonable in this particular instance, it’s important to remember that we only have 10 point counts. If our writeup is “effect of pasture conversion between 3000-3500m in the Cundinamarca department of Colombia”, that might be okay, but if we are interested in the effect of pasture conversion in the Colombian Andes in general, it would be hard to argue that 10 points is a representative sample of cloud forest.